Fixed #7. 1)Disable the multi-thread and 2) Modified kernel codes to avoid unloop in axpy function when incx==0 or incy==0.

interface/axpy.c

@@ -81,6 +81,11 @@ void CNAME(blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint inc
 #ifdef SMP
   nthreads = num_cpu_avail(1);
 
+  //disable multi-thread when incx==0 or incy==0
+  //In that case, the threads would be dependent.
+  if (incx == 0 || incy == 0)
+	  nthreads = 1;
+
   if (nthreads == 1) {
 #endif
 

interface/zaxpy.c

@@ -83,6 +83,11 @@ void CNAME(blasint n, FLOAT *ALPHA, FLOAT *x, blasint incx, FLOAT *y, blasint in
 #ifdef SMP
   nthreads = num_cpu_avail(1);
 
+  //disable multi-thread when incx==0 or incy==0
+  //In that case, the threads would be dependent.
+  if (incx == 0 || incy == 0)
+	  nthreads = 1;
+
   if (nthreads == 1) {
 #endif
 

kernel/x86_64/axpy_sse.S

@@ -1463,6 +1463,12 @@
 .L50:
 	movq	M, %rax
 	movq	Y, YY
+//If incx==0 || incy==0, avoid unloop.
+	cmpq	$0, INCX
+	je  .L56
+	cmpq	$0, INCY
+	je  .L56
+
 	sarq	$3,   %rax
 	jle	.L55
 	ALIGN_3

kernel/x86_64/axpy_sse2.S

@@ -805,6 +805,12 @@
 .L40:
 	movq	Y, YY
 	movq	M,  %rax
+//If incx==0 || incy==0, avoid unloop.
+	cmpq	$0, INCX
+	je  .L46
+	cmpq	$0, INCY
+	je  .L46
+		
 	sarq	$3, %rax
 	jle	.L45
 	ALIGN_3

kernel/x86_64/zaxpy_sse.S

@@ -2893,6 +2893,12 @@
 	unpcklps %xmm13, %xmm15
 #endif
 
+//If incx==0 || incy==0, avoid unloop and jump to end.
+	cmpq	$0, INCX
+	je  .L200
+	cmpq	$0, INCY
+	je	.L200
+
 	movq	Y, YY
 
 	movq	M,  %rax
@@ -3105,8 +3111,42 @@
 	addps	%xmm1, %xmm8
 
 	movsd	%xmm8,   (Y)
+	jmp	.L999
 	ALIGN_3
+		
+.L200:
+    movq    M, %rax
+	cmpq	$0, %rax
+    jle .L999
+    ALIGN_3
+
+.L201:
+    movsd    (X), %xmm0
+    addq    INCX, X
+
+#ifdef HAVE_SSE3
+    movshdup %xmm0, %xmm1
+    movsldup %xmm0, %xmm0
+#else
+    pshufd  $0xf5, %xmm0, %xmm1
+    shufps  $0xa0, %xmm0, %xmm0
+#endif
+
+    mulps   %xmm14, %xmm0
+    mulps   %xmm15, %xmm1
+
+    movsd    (Y), %xmm8
+
+    addps   %xmm0, %xmm8
+    addps   %xmm1, %xmm8
 
+	movsd   %xmm8,   (Y)
+    addq    INCY, Y
+		
+	decq    %rax
+    jg  .L201
+    ALIGN_3
+		
 .L999:
 	xorq	%rax, %rax
 

kernel/x86_64/zaxpy_sse2.S

@@ -1416,6 +1416,12 @@
 
 	movq	Y, YY
 	movq	M,  %rax
+//If incx==0 || incy==0, avoid unloop and jump to end.
+	cmpq	$0, INCX
+	je  .L58
+	cmpq	$0, INCY
+	je	.L58
+		
 	sarq	$3, %rax
 	jle	.L55
 
@@ -1769,6 +1775,7 @@
 	andq	$1, %rax
 	jle	.L999
 
+.L58:	
 	MOVDDUP( 0 * SIZE, X, %xmm0)
 	MOVDDUP( 1 * SIZE, X, %xmm1)
 
@@ -1781,6 +1788,9 @@
 
 	movlpd	%xmm8,   0 * SIZE(YY)
 	movhpd	%xmm8,   1 * SIZE(YY)
+		
+	decq %rax
+	jg	.L58
 	ALIGN_3
 
 .L999: